Here's what we're building. Audit the code, test it, enjoy. Local AI 1. detects your hardware 2. tries to find an exl3 optimised recipe 3. downloads, runs the model 4. pick one of 8 agents, pi/omp/codex/claude/opencode/etc.. 5. runs the agent

Any chance of Intel iGPU support? Chips like the Core Ultra 5 series have 4 Xe-cores, shared RAM and can already run 3B–8B Q4 models via llama.cpp Vulkan/SYCL. They are not fast, but these laptops will run ause able local LLM depending on ram size. This would cover a lot of machines.

I already added intel arc b70 support, you add support by putting recipes in here give it to any clanker and it'll figure it out

The real unlock isn’t just running AI locally it’s removing the friction between hardware and useful work. Detect → optimize → download → run → build. If you nail that UX, local AI stops feeling like infrastructure and starts feeling like a product.

Love to see it 🤘 I submitted a PR to get @kujolang installed by default to help make it easier for developing with AI. I think it'd be a great addition alongside all of the local first work you're doing 🫡

How it's different from ODS? Osmantic AI?

It doesn’t

I can help you beat that 5090 tps for 32GB

can i make a few formal requests, now that your are officially in the project (congrats, btw. nice job), to include a few things here for local ai support: - a unified gateway covering both my local models, cloud APIs, and subscriptions (@opencode go and @openai codex) [i've got a private version of this, lmk if you want access to the current codebase] - @dee_hw and @autonomous_labs have some awesome stuff w/ (omagrid); i would like to please be able to discover/add other gpus on my network - local classifier models that are integrated for blocking sensitive data from going out to 3rd party providers (openrouter, openai, etc), probably will need tooling for helping users configure what they care about, possibly even enabling fine tuning?

I saw what you are doing with Omarchy now. Amazing news. I was thinking about reaching out to present my agent framework when it's released. It could be an amazing fit. Hopefully we can talk then. Congrats!

Congrats on the @OmarchyLinux sponsorship boy. Got a long way to go. Chill and be the best dad 💯🔥

Cooking something along those lines too.

Est-ce que tu continues le projet d'adaptation d'Omarchy sur les puces M3 et M4

Create an IPFS LIBRARY OF "CERTIFIED" LLMs so we can download and install

Yo @grok what how does exl3 help? Isnt there severe quality loss at such compression?

Any chance for me to contribute with a 3050 😅? Seriously asking, not EXL3s though, custom models with custom llama.cpp patches for them. Yes or No?

Can Hermes be added to the list of agents ?

amd too? 7900 or r9700?

Sei un buon investimento, grazie

Is it working for the dual GPU set up with tensor parallel?

@grok add this to my reminder list and ping me in 3 months 😊 about the details what it is now and what it will become then 🥳

Where’s DSH?